High Precision Leaf Instance Segmentation for Phenotyping in Point Clouds Obtained Under Real Field Conditions

In-Field Phenotyping Based on Crop Leaf and Plant Instance Segmentation

In-Field Phenotyping Based on Crop Leaf and Plant Instance Segmentation

There are maize phenotyping methods including organ segmentation, leaf traits extraction, etc.

Using state-of-the-art object detection and image segmentation techniques we are able to infer the yield on different plant experiments. We demonstrate and produce an effective proof of concept computer vision model employing transfer learning that is capable of accurate predictions on a variety of Tomato and Basil plant species across 4 different experimental datasets. Finally, we visualize our yield predictions and results in an interactive dashboard combining the predictions with the operational experiment data (sensor data), showing the accuracy of the predictions and providing a solid base for future work.

My friends Nikhil Vijay S, Harish Kumar Patidar and I did work on the leaf segmentation challenge, Mentored by Mr.Mohit Agarwal on an internship at Bennett University. In this project, we investigate the problem of segmenting rosette leaves from an RGB image, an important task in plant phenotyping. We propose a data-driven approach for this task generalized over different plant species and imaging setups. To accomplish this task, we use state-of- the-art deep learning architectures: UNET, a convolutional neural network for initial segmentation. Evaluation is performed on the leaf segmentation challenge dataset at CVPPP-2017. Despite the small number of training samples in this dataset, as compared to typical deep learning image sets, we obtain satisfactory performance on segmenting leaves from the background as a whole and counting the number of leaves using simple data augmentation strategies. Comparative analysis is provided against methods evaluated on the previous competition datasets.

Code fro the paper "Top leaf segmentation and phenotyping with RGB and LI-Cor data for Oat plants"

Automated plant phenotyping using SAM for per-plant trait extraction. Handles leaf overlap through pot-anchored segmentation. Zero-shot, small dataset friendly, biologically correct measurements.

U-Net implementation for vegetation and leaf segmentation using the CWFID crop/weed dataset. This project reproduces the segmentation pipeline from the related research paper and demonstrates deep-learning methods for precision agriculture and plant phenotyping.